Statistical modeling of proteome expression data in Manila clam, Ruditapes philipinarum, exposed to citrate capped gold nanoparticles (AuNP), as a model contaminant of environmental nanoparticle contamination

Trabajo presentado en el Scotland and Valencia workshop on Bayesian Statistics (ScoVa16 Workshop, VABAR), celebrado en Valencia el 28 y 29 de enero de 2016.Among others, proteomic research aims to identify and quantify relative changes in pro-tein abundance between individuals under different circumstances. In ecotoxicology, pro-teomics is applied to test organisms exposed to environmental contaminants, and is es-pecially useful for the evaluation of chronic low-level exposure effects that manifest atmolecular levels of organization. Obtained information can be used to better understandmetabolic pathways and networks the contaminant interferes with, providing a betterknowledge of the mode of action of the contaminant, as well as for contaminant specificbiomarker development for environmental monitoring purposes. In a laboratory-basedexperiment, we exposed the non-target marine bivalve, the Manila clamRuditapes philip-pinarumto an environmentally relevant concentration (0.75μg L−1) of weakly agglomer-ating citrate AuNPs (∼20 nm). An 8-plex iTRAQ-based bottom-up proteomic approachfollowed by tandem mass spectrometry led to the identification and quantitation of 2200expressed proteins. In order to identify statistically significant features, the obtained datawas analysed using multiple contrast analysis corrected by FDR aimed at the control ofthe proportion of false positives among rejected hypotheses. However, the experimentaldesign applied resulted in a higher number of variables than samples or observations lead-ing to an over-conservative approach. Alternatively, the restricted model regressions Lassoand Elastic-Net were applied to provide better information on what proteins differentiatebetween control and treated clams through penalization of the coefficients of the variablesthat are not associated with the response variable zero. Elastic-Net revealed 105 iden-tified proteins of which many were the same as those obtained using multiple contrasts.Ultimately, samples were classified after dimension reduction by PLS-DA. This provideda good classification of the samples, in which all variables are projected along the com-ponents creating the model. Several obtained proteins were related to oxidative stress,inflammatory response and cytoskeleton and correlated with several features identified inq-PCR gene expression analysis carried out on the same samples previously.Peer reviewe